CAJ | 학술논문

为建立某一飞行科目的模板样本集，提出一套基于神经网络和卡尔曼滤波的数据处理方法。任选一次试飞样本，建立Kohonen自组织神经网络进行参数降维、聚类分析、特征提取等，使样本量缩减90％以上，得到该科目的模板样本集。用处理后的样本训练BP神经网络，对其他未经处理的试飞样本进行载荷预测，误差均在3％之内，说明处理后的样本能代表该科目的数据特点，即为模板样本集。方法可以为飞机载荷监控数据库的完善工作服务。
위건립모일비행과목적모판양본집，제출일투기우신경망락화잡이만려파적수거처리방법。임선일차시비양본，건립Kohonen자조직신경망락진행삼수강유、취류분석、특정제취등，사양본량축감90％이상，득도해과목적모판양본집。용처리후적양본훈련BP신경망락，대기타미경처리적시비양본진행재하예측，오차균재3％지내，설명처리후적양본능대표해과목적수거특점，즉위모판양본집。방법가이위비궤재하감공수거고적완선공작복무。
In order to establish the sample template of a flight course ,a data processing methodology based on Neural Networks and Kalman Filter is provided .For the sample set of an arbitrarily selected flight test in a course ,the Kohonen Self-organized Feature Maps are established to achieve parametric dimension reduction ,data clustering and feature extraction , resulting in over 90% of the original samples canceled and yielding the sample template of this flight course .The accuracy check begins by the establishment of BP Neural Networks trained by the processed sample set ,where the loads of other unprocessed sample sets in the same course are predicted .The prediction error can be within 3%, indicating that the processed sample template well represents the feature of this course and is therefore proved to be the sample tem-plate.The methodology presented above can assist help to the perfection of the database of aircraft struc -tural loads monitoring .